Schneier on Security

Clive Robinson • July 20, 2025 11:04 AM

@ ALL,

The tool of tools, has toolmarks that lead blame home.

For those that do not know “Ghost Gun” referres to the production of a fire arm or other projectile weapon that does not have a required serial number.

In the US this is usually the sub or under frame called the “receiver”[1] that holds the trigger mechanism etc. In other countries it might be the breach and barrel that get marked as with a “test fire” record means any bullet fired through it is directly tracable.

The thing is that much of a fire arm can be made of things other than metal. Thus could be produced on a 3D Printer which is what all the noise in the MSM about “Ghost Guns” is actually all about.

The thing is that be it an extrusion process or a routing process or even an ultra modern laser cutter they all have “beds” on which the piece being made is constructed, and an XYZ mechanism that “moves the tool head”.

Like all things mechanical the XYZ mechanism and other moving parts in these 3D devices all have “mechanical slop”.

That is if you move the tool head from point A to point B then back to point A it will not quite be the original point A but A’.

Whilst the difference between points A and A’ might be microscopically small it is measurable. Not just in the tool head movement but in it’s cut tracks on the piece being made. Thus the basis for a “fingerprint” measurement that used to be used on typed pages to identify the typewriter or later printer.

Well one US Police Forensic Investigator Kirk Garrison thinks that these residual tool marks can be used as an identifying tool mark.

The story originally appeared on 404 Media but they in effect have the equivalent of a PayWall up and controlling access (see https://www.404media.co/3d-printing-patterns-might-make-ghost-guns-more-traceable-than-we-thought/ ).

The story can also be found in other places such as,

https://www.tomshardware.com/3d-printing/police-link-3d-printers-to-ghost-guns-using-fingerprints-from-printers-toolmarks-left-behind-during-printing-can-make-ghost-guns-traceable.

The important thing to note though, is that our ability to measure accurately at low cost, currently out strips our ability to manufacture at or beyond those tolerances by low cost machinery. And personally I can not see this “measure over manufacture” changing any time soon.

But it is also something that goes back in time quite a while, as I’ve mentioned before to Locard’s exchange principle[2],

“Every contact leaves a trace :

Wherever he steps, whatever he touches, whatever he leaves, even unconsciously, will serve as a silent witness against him.

Not only his fingerprints or his footprints, but his hair, the fibres from his clothes, the glass he breaks, the tool mark he leaves, the paint he scratches, the blood or semen he deposits or collects. All of these and more, bear mute witness against him. This is evidence that does not forget. It is not confused by the excitement of the moment. It is not absent because human witnesses are. It is factual evidence. Physical evidence cannot be wrong, it cannot perjure itself, it cannot be wholly absent. Only human failure to find it, study and understand it, can diminish its value.”

Back then “tool marks” would be exampled by say the impression the jagged claw of a crow-bar left on the paint and wood of a door or window frame. But the principle remains the same, if it’s measurable and repeatable thus testable then it’s very probably admissable as evidence.

The thing is whilst it’s usually considered for “tangible physical objects” it can also apply to the substrate of “intangible information objects” like software and data files.

So if you are planning a life of crime, remember that Locard’s Principle probably holds true for anything you do…

[1] The use of the term “receiver” goes back to US legislation in the 1960’s. When making a firearm for your own personal use from blanks through to a kit of parts was both legal and did not require registration etc. But times change,

https://en.m.wikipedia.org/wiki/Receiver_(firearms)

[2] Whilst generally true “it cannot perjure itself” needs to be treated with caution. The fact an item is found at a crime scene with a drop of your blood on it, does not mean that you were ever at the crime scene or at the time the crime was being committed. That is the object with “your spoor on it” could have been introduced to the crime scene by another agent or the tests inaccurate for various reasons. As I found out when not even 10 you could fake human fingerprints fairly easily (I much later found out that the “Sherlock Holmes” author was well aware of it nearly a century before).

Clive Robinson • July 21, 2025 6:15 PM

@ Bruce, not important, lurker, ALL,

If current AI does not scale and burns the world as it churns, then what?

Some who think AI is unnecessary would say

“Give it up as a bad lot.”

Others who think AI might open up ‘new ways’ might say,

“Find a better way.”

Me personally see Current AI as,

“A bit pants or meh”

Because I’ve done work in robotics driven by various forms of AI off and on since the 1980’s (and you get to recognise the “give me money” / “ego research” signs).

In a way I look at most “Current AI” as what it actually is “under the hood”

“A statistical tool of sequentially applied rules.”

Which kind of points out why it can not effectively scale in it’s current “under the hood” instantiations.

So how to climb out of the money pit of “ever decreasing returns”?

One way out is to come up with a way to “efficiently go parallel” but the way that is being done has it’s limits as DNN’s implicitly show.

Another is the oft self defeating “do more with less” that is reduce the number of sequential rules by making each rule do more with less. It fails under that MBA notion of “just increase productivity to increase profit”.

There is however another way of getting the needed improvement which is “preselect the input”. But it smacks to many of 1980’s “Expert Systems” (which is as far as I can see the only form of AI that actually shows useful returns).

There are other “engineering ways” but few want to go down those routes for various reasons[1].

But “the current way” is in effect hard limited by “C the speed of light” and “Inefficiency heat death”. Yes there are ways to reduce the distance between nodes but each node needs a minimum amount of energy thus the energy density goes up rather rapidly and meltdown of thermal runaway and similar becomes a very real issue.

So the argument is make the nodes and interconnects smaller and way more efficient to cut waste is also a way to go[2]… But the laws of nature put limits on most of that… As for the rest it’s mostly “Blue Sky Forever Research” as evidenced by “Quantum Computing”.

So these are your base options and they are not going anywhere fast (or at all).

That said there are always Hybrid Solutions. That can be used as useful stepping stones. The ideas behind this can be seen in,

https://techxplore.com/news/2025-07-democratizing-ai-powered-sentiment-analysis.html

You can see what they are doing in the “Our approach” section,

“First, we convert each input sentence into a fixed-length vector by mean-pooling over its token embeddings. This transforms text into a semantically rich representation without the overhead of token-by-token processing. Next, we fine-tune these sentence transformers directly on labeled sentiment data.”

Note “fixed-length”, this puts a constraint not just on the “input” but also increases the gaps in the vector space. Which has advantages and disadvantages[3]. The most obvious being “averaging out the gaps” to try to turn “spiky into similar clumps” on one level, then differentiate by chosen differences on another level,

“By applying supervised loss functions, CosineSimilarity and CoSENT to align pairs of same-sentiment sentences, SoftmaxLoss to sharpen class boundaries, and variants of triplet loss (BatchAll, BatchHard, SoftMargin, SemiHard) to push dissimilar sentiments apart—we sculpt the embedding space so that positive, neutral and negative examples naturally cluster.”

In effect they are both “limiting” and “preselecting” the input” so that they are “pushing square pegs into round holes”.

Thus they make certain types of gain and / or cost reductions.

I expect to see a lot more “hybridisation to happen” in the near future as a way to,

1, Make AI more secure / surveillance reduced.
2, Make it less generic thus more useful as a tool.
3, Increase performance without the need for specialised engineering.
4, Reduce energy and cooling costs.

And a few other things that some might consider beneficial such as,

5, Keep the AI hype going.

The reality is it’s a move from the woo woo AGI notion back towards functional “Expert Systems” and the potential to “be actually useful” thus potentially earn actual income over expenditure.

[1] A classical example of this is the history of early nuclear fuel enrichment and why we ended up with cascades of inefficient centrifuges in what look like infinite feedback loops as the only practical way forward.

[2] In theory you can get the energy back in a useful way by using logic elements that are reversible like Toffoli / Fredkin / Deutsch gates. In effect to worm around the “AND is non reversible” issue by creating a cascade of gates where information is not lost.

https://ieeexplore.ieee.org/document/10218349

Combine this with “fuzzy elements” and you get this line of thinking,

https://slideplayer.com/slide/16811277/

(There are other papers but nearly all are behind “Kings Ransom” paywalls).

[3] A friend –nolonger with us– had a sarcastic expression,

“A hammer works better when you make screws and bolts more like nails.”

But there is a germ of hybrid in there… Some nails now have screw like features, because it strengthens their hold as well as making removal still possible. It is a “hybrid solution” that works well for fixing certain types of materials (like fiber based composites and soft woods).

Clive Robinson • July 22, 2025 6:09 PM

@ Bruce, ALL,

US Sherif uses “smart meter” readings to send goons in.

I’ve mentioned on a few occasions over the years that “Smart Meters” could spy on what goes on inside your home but appear “custom made” for such surveilling.

Well,

https://arstechnica.com/tech-policy/2025/07/eff-moves-to-stop-power-utility-reporting-suspected-pot-growers-to-cops/

“According to a motion the Electronic Frontier Foundation filed in Sacramento Superior Court last week, Nguyen and Decker are only two of more than 33,000 Sacramento-area people who have been flagged to the Sheriff’s department by the Sacramento Municipal Utility District, the electricity provider for the region. SMUD called the customers out for using what it and department investigators said were suspiciously high amounts of electricity indicative of illegal cannabis farming.

The EFF, citing investigator and SMUD records, said the utility unilaterally analyzes customers’ electricity usage in “painstakingly” detailed increments of every 15 minutes. When analysts identify patterns they deem likely signs of illegal grows, they notify Sheriff’s investigators. The EFF said the practice violates privacy protections guaranteed by the federal and California governments and is seeking a court order barring the warrantless disclosures.”

This illegal searching has been going on for atleast five years…

And unsurprisingly it’s littered with false positives…

At some point some gun totting numpty is going to shoot someone…

As for the results of all these illegal search attempts, it would be interesting to find out just how many are “a waste of police resources” that the citizens are funding through taxes…

Clive Robinson • July 23, 2025 6:15 PM

@ ALL,

Developer is getting the NEW AI look and feel or not.

There has been as we know a “kick-back” over the forcing of “AI In Everything” with some saying they will boycott and some actually boycotting upgrades with forced AI in them.

In some cases for very “Good and Proper Reasons” involving security, privacy, and legal obligations. Not least because “Current AI” is used as a form of “Client Side Scanning” that does an “ET Phone home to the mothership” with everything it can gobble up. Another being it can in Agentic Form destroy entire code bases and tens of thousands of hours of actual human hard graft work.

So at the very least “Current AI” is not in a “road safe condition” when you need the wheels to stay on and the breaks to be reliable so you actually get to your chosen destination.

Yet others want to throw caution to the wind and “get with the grove” as was said in the 1960’s or in more modern parlance “feel the vibe”. In either time line it is a mind altering experiment where “Caution is advised”…

So what to do as a developer?

Well Zed Industries say,

‘Our goal at Zed has always been to build the world’s best code editor.’

https://zed.dev/blog/disable-ai-features

Which puts it “front and center” in a developers tool chain and represents one of those “critical points of failure” that can make or break a project.

So they need to “grasp the thistle” on this issue both early and decisively.

Which is why they say,

‘For many, the best editing experience must include first-class agentic AI. That’s why we’ve made Zed the world’s fastest AI code editor and launched the Agentic Engineering series to explore how programmers can use AI effectively.’

And as importantly they go on to say,

‘But we’ve heard from others on GitHub discussions and issues who either cannot or would prefer not to use Zed’s AI features.’

Thus they admit up front there is a very real issue that needs to be addressed as you can not be both at the same time…

So their solution,

‘That means giving you control over your development environment, including the choice to work without AI if that’s what works best for you.’

So they offer four basic options,

‘You’re now able [to] add a global setting to disable Zed AI [in] your settings.json file.’

‘Zed AI service[:] your code and prompts are discarded after each request, never stored persistently, and never used for training. We also maintain zero-retention agreements with Anthropic to ensure your code stays private.’

‘Bring your own keys: Easily connect to AI providers you trust using your own API keys, giving you direct control over the vendor relationship.

Keep it local: Zed also supports local AI models that keep your code completely on your machine, so nothing ever leaves your development environment.’

Which begs the question,

“What is so difficult about adding such options?”

The answer is mostly “It’s not” so why do the Silicon Valley Mega Corps want you to think it’s impossible and that you “Must have AI”.

The answer of course is that,

ET phone home to the mother ship

The Mega Corps want to steal every key click yo make so they can commoditize not just your work but you yourself.

Because there is no other way for them to make a return on the so far billions/trillions of sunk costs of “Current AI LLM and ML systems”.

The Mega-Corp view is,

“We can sell you into slavery to hide our stupidity.”

Hence their ramming of their work stealing AI down your throat.

The thing is all the Mega-Corps jumped into “Current AI” blindly and have taken with it a sizable part of not just the US economy but the first world economy.

The last time “to big to fail” did the blind “We’re jumping in because they’re jumping” was Sub-Prime Mortgages, and we know the economic crater that created because we are still trying to crawl out of it.

The Silicon Valley Mega-Corps have in fact made a pact… The only question is what type,

“With the Devil, Suicide, or both?”

When commentators make potentially controversial arguments they tend to do one of two things. Say “let me know in the comments below” or they “turn comments off”. All I ask is that people think about it and what it means to them and their future, because I’m pro “giving people choice” even when they want to,

“Shoot themselves in the foot in the life boat and sink the rest of us with them.”

Clive Robinson • July 25, 2025 8:27 AM

@ Bruce, ALL,

Political discrimination by algorithm wrecking lives.

I’ve been hearing stories / reports about the Dutch Childcare Scandle that discriminated against people in line with the political mantra being pushed by politicians

Some reports are blaming AI, but I’ve not been able to find verifiable evidence.

The fact that the racist algorithm got traced back far enough for many Dutch Politicians to resign rather than be held responsible / accountable, further suggests it was a deliberate implemented by humans algorithm, rather than bassed on Machine Learning.

The problem is the muddying with AI etc is allowing this scandal to be not just “arms lengthed” but also “swept under the rug” with expressions like,

“xenophobic algorithm”

Being bandied about, which in many minds sounds like “the rise of the machines” or “AGI” thus three basic things happen,

1, Conflation muddies the water to the benefit of those responsible.
2, Those responsible retain power and carry on.
3, The view of AI gets distorted probably irrevocably in many minds.

In this case whilst the Dutch “cabinet” has resigned they still remain in power and will re-stand in a General Election and due to the sentiments raised in the Dutch Citizenry will in high probability be re-elected…

So empowered to do “more of the same” but… With a little less obviousness, which is where “Current AI LLM and ML Systems” will almost certainly be considered if not used in some way.

I’ve been warning about this sort of “Political mantra” codified into automated systems for some time now, and more importantly how what are claimed as “black box systems” will get increasingly used to deflect accountability and responsibility. The DNN’s in Current AI LLMs are claimed to be “black box” likewise the Current AI ML systems that extract the statistics that get built in. Thus the fuel is piled high onto basic human rights and the current political climates driven by “Mantra” around the world are striking matches.

I honestly can not see this ending well as you can not legislate against bad algorithms when the algorithms can not be seen or reasoned about and would mostly make no sense in human eyes.

Any way two articles on this Dutch Child Care Scandal that appear about as unbiased as you can get when the information about the scandal is apparently being mired up under authoritarian Political leadership diktat,

https://www.vice.com/en/article/how-a-discriminatory-algorithm-wrongly-accused-thousands-of-families-of-fraud/

https://www.amnesty.org/en/latest/news/2021/10/xenophobic-machines-dutch-child-benefit-scandal/